Yann LeCun, the AI pioneer betting a billion dollars that chatbots are a dead end

Yann LeCun

Photo: Ecole polytechnique from Paris / CC BY-SA 2.0, via Wikimedia Commons

Born	July 8, 1959 Paris
Occupation	Software Engineer
Awards	PAMI Distinguished Researcher Award · Harold Pender Award · Turing Award

For years Yann LeCun sat inside one of the companies racing to build ever larger chatbots and told anyone who would listen that the approach would never reach human-level intelligence. Now he has left to prove it, with a little over a billion dollars of other people’s money. AMI Labs, the Paris company he co-founded after walking out of Meta, is building what he calls world models: systems that learn how the physical world behaves from video and sensor data instead of from text. The scientist whose work on neural networks made the current boom possible is staking his reputation on the claim that the boom’s central technology is a detour.

His family name was originally written Le Cun and comes from the old Breton Le Cunff, from the country around Guingamp in northern Brittany. He was born in Soisy-sous-Montmorency, in the suburbs north of Paris, trained as an engineer at ESIEE Paris and took a PhD in computer science at Université Pierre et Marie Curie. His thesis proposed an early form of backpropagation, the procedure that lets a neural network correct itself from its errors and that sits under almost every system now described as artificial intelligence. He then spent a year in Toronto as a postdoctoral researcher under Geoffrey Hinton.

In 1988, a year after his doctorate, he joined AT&T Bell Laboratories in Holmdel, New Jersey. There he built the convolutional neural networks known as LeNet, which learned to read handwritten digits by looking at examples. A check-reading system he helped develop was widely deployed by NCR and other companies. At AT&T Labs-Research, which he joined in 1996 to head image processing, he worked with Léon Bottou and Patrick Haffner on DjVu, a compression format for scanned documents that the Internet Archive still uses to serve digitised books.

Neural networks then fell out of fashion, and LeCun kept working on them anyway. He moved to New York University in 2003, where he still holds the Jacob T. Schwartz chair at the Courant Institute, became founding director of its Center for Data Science in 2012 and the following year co-founded the ICLR conference with Yoshua Bengio. When deep learning came back, it came back on the ideas he, Bengio and Hinton had defended through the lean years. The three men shared the 2018 Turing Award, computing’s highest prize, which is why newspapers still call them the godfathers of AI. By then LeCun had been working for Facebook for five years: in December 2013 the company hired him to set up its AI research lab, FAIR.

FAIR became one of the most open laboratories in the industry, publishing its papers and code, and LeCun became the industry’s loudest advocate of releasing model weights so that anyone could build on them. The lab trained the first LLaMA, released to researchers in February 2023. It also produced Galactica, a science-writing model whose public demo Meta pulled within about 48 hours in November 2022, after users showed it inventing confident nonsense. ChatGPT arrived two weeks later and the public loved it; LeCun has since said the world simply was not ready for Galactica. By 2025 Zuckerberg had put Meta’s AI effort under Alexandr Wang and a new superintelligence division, LeCun reported to Wang, and FAIR was hit by the October layoffs. In November he announced he would leave at the end of the year.

The difficulty with LeCun is that he is often right about the long game and combative about the short one. He has argued for years that large language models cannot reason or plan, while chatbots built on them became the industry’s main business. Weeks after leaving, he told the Financial Times that Meta’s Llama 4 benchmark results “were fudged a little bit”, with different versions of the model used for different tests, a charge Meta’s generative AI chief Ahmad Al-Dahle had denied when it first surfaced. On risk he stands apart from his fellow laureates: Hinton and Bengio now warn that AI could threaten humanity, while LeCun told Fortune in October 2026 that he is not worried about extinction at all, called effective altruism “super toxic” and said Anthropic‘s chief executive, Dario Amodei, is “completely deluded”. Warning that AI is too dangerous to put in the public’s hands, he argues, is regulatory capture. Critics answer that his optimism rests heavily on trust in the people who deploy these systems.

AMI Labs is the bet that settles the argument one way or the other. Unveiled in March 2026, it raised $1.03 billion at a pre-money valuation of $3.5 billion, which LeCun called one of the largest seed rounds ever. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions, with Nvidia, Samsung and Toyota Ventures among the other backers. Alexandre LeBrun, who ran the health start-up Nabla, is chief executive; LeCun is executive chair and keeps teaching at NYU. The company has its headquarters in Paris, teams in New York, Montreal and Singapore and about 60 employees, and builds its models on JEPA, the joint embedding predictive architecture LeCun has promoted for years, aimed first at industrial uses such as robotics and anomaly detection. LeBrun has said commercial applications could take years. In January 2026 LeCun also became founding chair of the technical research board of Logical Intelligence, a company working on energy-based reasoning systems.

Away from the lab, LeCun is a keen amateur astronomer: the image behind his post announcing AMI was the Veil Nebula, photographed from his own backyard. He is a French and American citizen, turned 66 in July 2026 and has three sons. He set out his own account of the field in Quand la machine apprend, published in France in 2019. The company name, he points out, is also the French word for friend.

AMI’s first product, still unnamed, is due soon, and LeCun has hinted that open models may be part of it. It will be the first public evidence of whether world models can do what he has promised for a decade. If it works, the man the industry kept describing as a contrarian will have been early again, as he was with neural networks. If it does not, his warnings about chatbots will keep their force, but they will come from someone who has had a billion dollars to build the alternative.

Tags: business, ceos, meta, business leaders, AMI Labs, Geoffrey Hinton, Llama, Yann LeCun

Related People

Related People

Discussion